Session 14 : Hybrid Quantitative Modeling

1. Session 14: Hybrid Quantitative Modeling approaches with application to health & safety PJ Giabbanelli, M Baniukiewicz, VS Pillutla, R Flarsheim

2. Logistics 11:15 – 12:30 Wewelcome questions during the demo Beyondthis talk… Want to learn more? Help yourself to a paper on safety or health! Want to collaborate? giabba@cs.niu.edu Giabbanelli et al. Hybrid quantitative modeling 2

3. Organization of this demo Whybother? How to decompose a problem so it can use multiple modeling techniques Our combination Usingfuzzy cognitive map (locally) and networks (globally) Doingit in practice Ilustrationthrough a software environment What’snext? Resources to put itinto practice Giabbanelli et al. Hybrid quantitative modeling 3

4. Why bother Giabbanelli et al. Hybrid quantitative modeling 4

5. Why bother PUBLISH NOW!! Giabbanelli et al. Hybrid quantitative modeling 5

6. Why bother Change state No change W. Rush, et al., Tech. rep NECSI 2003 Unhealthy Healthy Giabbanelli et al. Hybrid quantitative modeling 6

7. Why bother Bahr et al. proposed a model of obesity in whichyoubecome obese if half of yourfriends are obese. D. Bahr, R. Browning, H. Wyatt, J. Hill, Exploiting social networks to mitigate the obesity epidemic, Obesity 17 (2009) 723-728 Giabbanelli et al. Hybrid quantitative modeling 7

8. Why bother But if wewant to model it ‘right’, we know thatit’s not the merepresence of overweight people thatmakesyou gain weight. Theydon’tsneezetheirweight on you! Giabbanelli et al. Hybrid quantitative modeling 8

9. Why bother Eating Exercising Giabbanelli et al. Hybrid quantitative modeling 9

10. Why bother +5 kg Individual context matters! • You tell me to cycle to work. Active travel is nice, but I live 50miles away! • I agree that healthy cooking is great, but I don’t have the money or knowledge for it! Giabbanelli et al. Hybrid quantitative modeling 10

11. Why bother You need a way to model: • Global dynamics. Social norms spread through a social network. Agent-based model. Network model. Cellular automaton. • Local context. What a norm might change in somebody’s life depends on a complex decision-making process, highly sensitive to one’s specific context. Fuzzy Cognitive Maps. System dynamics. Giabbanelli et al. Hybrid quantitative modeling 11

12. Our combination Giabbanelli et al. Hybrid quantitative modeling 12

13. Our combination Network between individuals, fuzzy cognitive map within individuals Lilian SES Exercise Martha Garnet Obesity SES Exercise SES Exercise Stigma Food intake Obesity Obesity Stigma Stigma Gender Food intake Food intake Gender Gender Giabbanelli et al. Hybrid quantitative modeling 13

14. Our combination Initialization Matching Simulation Nate Philippe Andrea Steven Sriram Giabbanelli et al. Hybrid quantitative modeling 14

15. Our combination Initialization Matching Simulation Nate Philippe Andrea Influencing Influenced Steven Sriram Giabbanelli et al. Hybrid quantitative modeling 15

16. Our combination Initialization Matching Simulation Nate Philippe Andrea Steven Sriram Giabbanelli et al. Hybrid quantitative modeling 16

17. Our combination Nate Exercise .2 Philippe Awareness of diabetes SES Exercise 1 .5 .8 Awareness of diabetes Diabetes SES .3 0 .7 0 1 .2 Obesity .8 Diabetes 0 Obesity .3 Giabbanelli et al. Hybrid quantitative modeling 17

18. Our combination Frequently Asked Questions and Whining (1) So now I have to develop a network model AND a fuzzy cognitive map? That’s a lot! There are lots of Fuzzy Cognitive Maps out there. Ask Steven Gray or I  And making a network model is fast. So you’ll spend time on combining. Giabbanelli et al. Hybrid quantitative modeling 18

19. Our combination Frequently Asked Questions and Whining (1) So now I have to develop a network model AND a fuzzy cognitive map? That’s a lot! There are lots of Fuzzy Cognitive Maps out there. Ask Steven Gray or I  And making a network model is fast. So you’ll spend time on combining. Giabbanelli et al. Hybrid quantitative modeling 19

20. Our combination Frequently Asked Questions and Whining B. Giles et al., Soc Sci Med 64(3), 562 (2007) Giabbanelli et al. Hybrid quantitative modeling 20

21. Our combination Frequently Asked Questions and Whining Giabbanelli et al. Hybrid quantitative modeling 21

22. Our combination Frequently Asked Questions and Whining (2) I model geographical systems, not social systems. So is that stuff really for me? Our guiding example was on social norms and individual context. But you can also have a geographical network with local context! Giabbanelli et al. Hybrid quantitative modeling 22

23. Our combination Frequently Asked Questions and Whining (2) I model geographical systems, not social systems. So is that stuff really for me? Our guiding example was on social norms and individual context. But you can also have a geographical network with local context! Help yourself to a copy! Giabbanelli et al. Hybrid quantitative modeling 23

24. Our combination Frequently Asked Questions and Whining (3) How much data do you actually need to run such a model? Intuitively, the more you want to represent in your model, and the more data you’ll need to initialize it / validate it. We’re talking about highly detailed models, so it gets data intensive. Giabbanelli et al. Hybrid quantitative modeling 24

25. Our combination You need data to get the weights of the FCM Frequently Asked Questions and Whining (3) How much data do you actually need to run such a model? Intuitively, the more you want to represent in your model, and the more data you’ll need to initialize it / validate it. We’re talking about highly detailed models, so it gets data intensive. We’ve developed techniques and software to do that automatically by sending in questionnaires Giabbanelli et al. Hybrid quantitative modeling 25

26. Our combination You also need data to initialize all concepts Frequently Asked Questions and Whining This is where you spend lots of time digging data. Help yourself to a copy! (3) How much data do you actually need to run such a model? Intuitively, the more you want to represent in your model, and the more data you’ll need to initialize it / validate it. We’re talking about highly detailed models, so it gets data intensive. Also be aware that values are often not independent! Giabbanelli et al. Hybrid quantitative modeling 26

27. Our combination Frequently Asked Questions and Whining (4) If I want to simulate a big population with complex decision making, it’s going to take lots of operations! How scalable is that? The algorithms parallelize well, so you could implement your model to run on a high-performance computing cluster. Giabbanelli et al. Hybrid quantitative modeling 27

28. Our combination Frequently Asked Questions and Whining Giabbanelli et al. Hybrid quantitative modeling 28

29. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 29

30. Doing it in practice There are several software packages to do hybrid modeling. AnyLogic can combine agent-based models and system dynamics – check out training great videos by Nate Osgood! We developed a software to combine networks and fuzzy cognitive maps. We can discuss collaborations and assist you with the implementation if you’d like. Giabbanelli et al. Hybrid quantitative modeling 30

31. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 31

32. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 32

33. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 33

34. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 34

35. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 35

36. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 36

37. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 37

38. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 38

39. Doing it in practice Giabbanelli et al. Hybrid quantitative modeling 39

40. What’s next? Giabbanelli et al. Hybrid quantitative modeling 40

41. What’s next? Learning more Giabbanelli et al. Hybrid quantitative modeling 41

42. What’s next? Learning more Giabbanelli et al. Hybrid quantitative modeling 42

43. What’s next? Putting it to practice Use AnyLogic to combine agent-based models and system dynamics. It’s based in Java and, while it features a drag-and-drop for model elements, you’ll have to code. or if you want to combine networks/agent-based with fuzzy cognitive maps (like we presented), talk to us. We can help you with model design and implementation. or Design models using a single paradigm (e.g. ABM) because you don’t have the time or data, but be aware of the big simplifications you may be making. Giabbanelli et al. Hybrid quantitative modeling 43

44. Questions? giabba@cs.niu.edu www.dachb.com Thanks to the students Giabbanelli et al. Hybrid quantitative modeling 44